Book a demo

{{ errorMessages.first_name }}
{{ errorMessages.last_name }}
{{ errorMessages.email }}
{{ errorMessages.company_name }}
{{ errorMessages.job_title }}
{{ errorMessages.case }}
{{ errorMessages.recaptcha }}

By submitting this form I confirm I have read and accepted Privacy policy

1
Share your contact details to get started
2
Book a demo for a specific time
>
State-of-the-art industrial anomaly detection powered by Topological Data Analysis

Process up to 50k images with 3 hours of GPU time per month for free

The problem we solve

What Slips Through Supervised Detection Costs You

Traditional supervised models rely on pre-trained data—leaving your production lines vulnerable to unseen defects. From engine misalignments to chip irregularities, unidentified anomalies disrupt efficiency, safety, and profitability

How it works

1. Train the Model

Start with as few as 100 images to train a reliable anomaly detection model, leveraging Topological Data Analysis for effective results.

2. Refine Your Dataset

(Optional) Easily improve accuracy by removing anomalous images from the training set, guided by the model’s initial highlights of irregularities.

3. Name Anomalies

Label and track detected anomalies for future reference while identifying new, unseen irregularities as they arise.

4. Apply the Model

Deploy the model in your environment with our exported ONNX format, seamlessly integrated via our Python API.

State-of-the-art technology

MVTec AD Detection

The MVTec Anomaly Detection (MVTec AD) dataset is a widely used benchmark dataset for unsupervised anomaly detection and localization in images. It is specifically designed to evaluate algorithms that detect defects in industrial and manufacturing settings.

MVTec AD Detection (added noise)

The MVTec Anomaly Detection (MVTec AD) dataset with introduced image-level noise and synthetic anomalies at both the image and feature levels using gradient ascent to further test the model.

Use cases

SEM Image Analysis for Chip Production (Intel)

For Intel, our platform uses Topological Data Analysis to analyze SEM images in real-time, catching defects—like microfractures or contamination—that supervised systems miss. New anomalies can be labeled and turned into detectable patterns, while multi-class classification groups trends like surface roughness for proactive control. The result: fewer escapes, higher yields, and a resilient production line.

Anomaly Detection in Aircraft Engines (Rolls-Royce PLC)

During routine inspections, high-resolution images captured inside aircraft engines provide critical insights into component health. Partnering with Rolls-Royce PLC, our platform analyzes these images using Topological Data Analysis to pinpoint anomalies—such as micro-cracks or wear—that often evade traditional methods. This collaboration has accelerated problem identification by at least three times, while revealing subtle defects previously undetected. By highlighting these issues early, we enable precise maintenance, reducing downtime and enhancing engine reliability for one of the world’s leading aerospace manufacturers.

Features

Detect irregularities in your visual data

AnomalyTDA conducts automated defect inspection on your images, pinpointing areas with potential issues.

Upload your images
Easily import images via your browser, from a local directory with a dedicated setup, or directly from cloud storage.
Train without labeled data
Run AnomalyTDA across all your images to automatically detect process deviations and highlight irregularities.

Train on our high-performance infrastructure — FREE

AnomalyTDA simplifies the training process by integrating large-scale image models with Topological Data Analysis, offering 3 hours FREE GPU time per month up on our infrastructure.

Optimized GPU Training
Automatically schedules GPU usage, manages queuing, and trains complex models efficiently. Supports both dedicated installations and cloud-based setups.
Smart resource management
AnomalyTDA dynamically activates GPU servers only during training and shuts them down afterward, ensuring no resources are wasted on idle machines.
Seamless cloud integration
Leverage dynamic GPU scaling across major cloud platforms, including AWS, Azure, and Google Cloud.

Refine anomaly detection with annotations

Users can classify detected deviations and mark certain patterns as non-anomalous, improving the model’s accuracy over time.

Tag detected anomalies
Annotate flagged anomalies for future reference and analysis.
Label normal patterns
Reclassify certain detected anomalies as legitimate patterns to enhance detection precision in future analyses.

Enhance model accuracy with active learning

With thousands of detected anomalies and numerous classifications, users need guidance in prioritizing the lowest-confidence predictions for optimal model improvement.

Kickstart with just 20 images
Begin anomaly annotation with a small dataset—typically, 20 labeled images are enough for initial predictions.
Smart active learning
The platform pinpoints low-confidence predictions, helping you focus on the most impactful corrections first.
Refine through iteration
With a few cycles of corrections, your model rapidly improves—achieving near-perfect accuracy in no time.

Advanced image classification

Beyond anomaly detection, our platform offers powerful multi-label image classification—applicable to everything from wafer map pattern recognition to equipment behavior analysis.

Multi-label image classification
A fast and robust classification system that can identify multiple legitimate categories within a single image.
Customizable classification models
Train and fine-tune models to recognize industry-specific patterns, adapting to your unique manufacturing and quality control needs.

Deploy anomaly models on edge devices

Download and run models in industrial environments with low-power edge devices—no GPU required.

Export in ONNX format
AnomalyTDA models can be saved in the widely supported ONNX format, ensuring seamless deployment on edge devices.
Lightweight Python integration
Use our fully featured yet lightweight Python library, easily installable via PyPI, to integrate anomaly detection into your workflow effortlessly.

Pricing

Starter

Ideal for small teams or individual deployments analyzing images
Files
50k
Seats
3
GPU Time
3 hours/month
Available models
Tiny, Small
Deployment
Cloud only
Integrations
None
Support
Email

Enterprise

Built for high-volume data and dedicated deployments
Files
10M+
Seats
Custom
GPU Time
Custom
Available models
Tiny, Small, Meduim, Large
Deployment
Cloud only, Docker Image for virtual private cloud (VPC)
Integrations
Storage Integration, VPC Integration, SSO/SAML
Support
Email, On-call support, Solution engineering hours

Build vs Buy

Internaly-built solutions

$400k+
in development cost
$120k / year
in maintenance
Talent Needs
Rare ML experts, hard and pricey to hire
Time Delay
12–18 months to launch, delay of ROI
Technical Debt
Growing complexity, future rework
Focus Drain
Shifts effort from core priorities
Not future-proof
Critical reliance on employees who developed the tool’s code

AnomalyTDA

State-of-the-art
Models for anomaly detection, object detetion and classification
On-prem or VPC deployment
Full control of your data
Engagement managers
To support you on every step
Model export in ONNX format
Run models on edge devices, no license restrictions or GPU needed
Designed for industrial usage
Top Results, No Coding Needed
Proven Results
Trusted by industry leaders like Intel and Rolls-Royce
Cost Predictability
Transparent pricing, no hidden maintenance fees

Let’s get in touch

Receive updates

Please input your contacts if you wish to receive AnomalyTDA updates.

We're here to help

We work closely with you to find out how AnomalyTDA may meet the needs of your organisation. Provide details about what you plan to achieve and we will contact you as soon as possible.